This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2986259, IEEE Access

SPECIAL SECTION ON MACHINE LEARNING DESIGNS, IMPLEMENTATIONS AND TECHNIQUES

Received January 16, 2020, accepted January 26, 2020, date of publication January 29, 2020, date of current version February 6, 2020.
Digital Object Identifier 10.1109/ACCESS.2020.2970178

A Novel Software Engineering Approach Toward

Using Machine Learning for Improving the
Efficiency of Health Systems
MOHAMMED MOREB 1, TAREQ ABED MOHAMMED2, OGUZ BAYAT1, AND OGUZ ATA1
1GraduateSchool of Science and Engineering, Altinbas University, 34217 Istanbul, Turkey
2Tareq Abed Mohammed, College of Computer Science and Information Technology, University of Kirkuk, Kirkuk 36001, Iraq

Corresponding author: Mohammed Moreb (mahammed.moreb@ogr.altinbas.edu.tr)

ABSTRACT Recently, machine learning has become a hot research topic. Therefore, this study investigates
the interaction between software engineering and machine learning within the context of health systems.
We proposed a novel framework for health informatics: the framework and methodology of software
engineering for machine learning in health informatics (SEMLHI). The SEMLHI framework includes four
modules (software, machine learning, machine learning algorithms, and health informatics data) that orga-
nize the tasks in the framework using a SEMLHI methodology, thereby enabling researchers and developers
to analyze health informatics software from an engineering perspective and providing developers with a new
road map for designing health applications with system functions and software implementations. Our novel
approach sheds light on its features and allows users to study and analyze the user requirements and determine
both the function of objects related to the system and the machine learning algorithms that must be applied to
the dataset. Our dataset used in this research consists of real data and was originally collected from a hospital
run by the Palestine government covering the last three years. The SEMLHI methodology includes seven
phases: designing, implementing, maintaining and defining workflows; structuring information; ensuring
security and privacy; performance testing and evaluation; and releasing the software applications.

INDEX TERMS Health dataset analysis, machine learning, methodology, software development manage-
ment, software engineering.

I. INTRODUCTION
The field of health informatics (HI) aims to provide a large-
scale linkage among disparate ideas. Normally, a healthcare
dataset is found to be incomplete and noisy; as a result,
reading data from dataset linkage traditionally fails within the
discipline of software engineering. Machine learning (ML)
is a rapidly maturing branch of computer science since it
can store data on a large scale. Many ML tools can be
used to analyze data and yield knowledge that can improve
the quality of work for both staff and doctors; however,
for developers, there is currently no methodology that can
be used. Regarding software engineering, there has been a
lack of approaches to evaluating which software engineering
tasks are better performed by automation and which require
human involvement or human-in-the-loop approaches [1].
Big data has many challenges regarding analysis challenges
The associate editor coordinating the review of this manuscript and
approving it for publication was Shadi Aljawarneh
for real-world big data [2], including OLAP mass data,
mass data protection, mass data survey and mass data
dissemination.
Recently, a set of frameworks have been used to develop
data analysis tools such as Win-CASE [3] and SAM [4].
The market has vast data analysis tools that can discover
interesting patterns and hidden relationships to support deci-
sion makers [5]. BKMR used the R package as a statisti-
cal approach on health effects to estimate the multivariable
exposure-response function [6].
Augmentor included the Python image library for aug-
mentation [7], while for the visualization of medical treat-
ment plans and patient data, CareVis was used [8], as it was
designed for this task. Other applications require a visual
interface using COQUITO [9]. For health-care data analytics,
the widely known 3P tools [10] were used. Many simple
applications, such as WEKA, which provided a GUI for many
machine learning algorithms [11], while Apache Spark was
. used for the cluster computing framework [12], are powerful

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/
VOLUME 8, 2020 23169

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2986259, IEEE Access
M. Moreb et al.: Novel Software Engineering Approach Toward Using Machine Learning for Improving the Efficiency
TABLE 1. Big data analytics tools according to the task.
systems that can used in various applications for solving prob-
lems using big data and machine learning [13]. Table 1 sum-
marizes the main tools used for big data in analytics according
with respect to the task. Software engineering for machine
learning applications (SEMLA) discusses the challenges,
new insights, and practical ideas regarding the engineering of
ML and artificial engineering (AI) [14]. NSGA-II proposed
algorithms for real-world applications that include more than
one objective function for enhancing performance in terms
of both diversity and convergence [15]. ML algorithms in
clinical genomics generally come in three main forms: super-
vised, unsupervised and semi-supervised [16]. Interflow sys-
tem requirement analysis (ISRA) has been used to determine
the system requirements [17].
Electronic healthcare (eHealth) frameworks have replaced
traditional medical frameworks to improve mobile health-
care (mHealth) and enable patient-to-physician and patient-
to-patient interactions to achieve improved healthcare and
quality of life (QoL) [18]. Big data and IoT have been
used for improving the efficiency of m-health systems by
predicting potential life-threatening conditions during the
early stages [19]. Intelligent IoT eHealth solutions enable
healthcare professionals to monitor health-related data con-
tinuously and provide real-time actionable insights used to
support decision making [20].
Machine learning is a field of software engineering that
frequently utilizes factual procedures to enable PCs to
‘‘learn’’ by using information from saved datasets. Unsu-
pervised or information mining focuses more on exploratory
information investigation and is known as learning supported
by data analytics. Patient laboratory test queue management
and wait time prediction are a challenging and complicated
job. Because each patient might require different phase oper-
ations (tasks), such as a check-up, various tests, e.g., a sugar
level test or blood test, X-rays or surgery, each task can
consider different medical tests, from 0 to N , for each patient
according to their condition.
In this article, based on a grounded theory method-
ology [21], the researchers proposed a novel methodol-
ogy, SEMLHI, in developing a framework by defining the
research problem and methodology for the developers. The
SEMLHI framework includes a theoretical framework to
23170
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
support research and design activities that incorporate exist-
ing knowledge. The SEMLHI framework was composed of
four components that help developers observe the health
application flow from the main module to submodules to run
and validate specific tasks. This enables multiple developers
to work on different modules of the application simultane-
ously. The SEMLHI framework supports the methodological
approach to conducting research on health informatics. It
also supports a structure that presents a common set of ML
terminology to use, compare, measure, and design software
systems in the area of health. This creates a space whereby
SE and ML experts can work on a specific methodological
approach to enable health informatics software development
teams to integrate the ML model lifecycle. Our methodology
was applicable to current systems or in the development of
new systems that use the ML module for current systems,
which can be used in regular updates to add data to the
system, to perform irregular updates and to add new features
such as new versions of ICD diagnosis codes, minor model
improvements for bug fixes, new functionalities required by
the client, and new hardware or architectural constraints.
II. METHODS AND DISCUSSION
Based on original data collected from a hospital run by the
Palestine government covering the past three years, first,
the data were validated, and all outliers were removed. Then,
the remaining data were analyzed using the developed frame-
work to compare ML techniques that predict test laboratory
results. Our proposed module was compared with three sys-
tems engineering methods: Vee, Agile and SEMLHI. The
results were used to implement the prototype system, which
requires a machine learning algorithm. After the develop-
ment phase, a questionnaire was delivered to the developer
to indicate the results of using the three methodologies.
The SEMLHI framework was composed of four compo-
nents: software, machine learning model, machine learning
algorithms, and health informatics data. The machine learn-
ing algorithm component uses five algorithms to evaluate
the accuracy of the machine learning models for various
components.
We used the original data as the selected dataset to develop
a patient prediction test laboratory result prediction model,
and the patient was required to perform more than one test.
In this article, we focus on helping patients and doctors com-
plete their treatment tasks by using predictable test results
based on the International Classification of Diseases [22]
(ICD-10) and helping hospitals save time and reduce effort
dedicated to medical testing. Using the SEMLHI framework,
realistic patient data were analyzed carefully and rigorously
based on important parameters such as age, start time, end
time, patient treatment, and detailed treatment content for
each task. We identified the laboratory tests required for
patients based on their conditions and the operations per-
formed during treatment. The patient data included only codi-
fied variables, including ICD-10 codes, procedure codes, and
medication orders, often reduced to smaller subsets.
VOLUME 8, 2020
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2986259, IEEE Access
M. Moreb et al.: Novel Software Engineering Approach Toward Using Machine Learning for Improving the Efficiency

FIGURE 1. Bed rate for patients by city.

A. DATASET AND PREPROCESSING

The application delivery of applied ML models in health-
care was often hampered by the existence of isolated prod-
uct deployments with poorly developed architectures and FIGURE 2. Main features used on the sample dataset.

limited or nonexistent maintenance plans. The ‘‘Translat-

ing Research into Agile Development’’ (TRIAD) method
presents a five-step method for designing a tailored EHR
tool [23].
The SEMLHI models and methodology were developed
by including new software systems connected to real datasets
and presented knowledge from the data using ML algorithms
to improve the efficiency of the required system. The dataset
case studies discussed in this article were set within the
context of Palestine hospitals and centers. Three hospitals and
nine medical centers were used for our dataset. Figure 1 illus-
trates the summary of bed rats per 1k patients distributed
across 12 cites. Furthermore, data collection was conducted
over the last three years, and 458k patients were identified
with corresponding patient nos. Overall, for the PMC dataset,
141k patients with 1.63% missing, a mean of 1.08M, a std dev
of 554k, a min of 10k, and a max of 1.04M were included.
For the age label, 141k patients with 1.63% missing, a mean
of 32.24, a std dev of 26.25, a min of 0, a max of 88, and a
median of 29 were considered.
B. AVAILABLE FEATURES OF PATIENTS
The patient dataset included 457914 cases and nine tables.
Each table had different features, and many techniques could
be implemented, such as semantic coordination for intelli-
gent databases [24], feature selection problems using genetic
algorithms [25], and new gene-weight mechanisms [26].
Some features were connected with other tables to build
datasets describing the main attribute, as mentioned in the
next section.
The laboratory test data include 200,000 cases (columns);
each case has a basic attribute such as the patient no., gender,
age, department, diagnosis code, description, and date of the
lab test. Figure 2 summarizes the main features used on the
sample dataset from all data, along with the distribution,
TABLE 2. Comparison of the three system engineering methods, Vee,
Agile and SEMLHI, proposed in this article.
center value and dispersion. The data are represented as
L= {l1, l2, l3, l4, . . . , ln}, where l is the item of the laboratory
test reports, and n is the number of items.
III. CONCEPTUAL FRAMEWORK AND DESIGN
A. SEMLHI METHODOLOGY
The SEMLHI methodology is used in software development
in the health area. For traditional applications, the devel-
opment process includes many methodologies, such as
the waterfall methodology, spiral methodology, and agile
methodology, which can be used to define and develop the
software. Table 2 illustrates the results of the comparison
between our methodology, the Vee [27] methodology, and the
Agile [28] methodology.
The SEMLHI framework methodology describes in detail
the process that was used when developing health software
and the mechanism used to integrate and use ML algorithms
with the development software. The SEMLHI methodology
provides a developer with a new road map for designing
health applications with system functions and software imple-
mentation. This framework includes ten stages, starting from

VOLUME 8, 2020 23171

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2986259, IEEE Access
M. Moreb et al.: Novel Software Engineering Approach Toward Using Machine Learning for Improving the Efficiency

FIGURE 3. SEMLHI framework users.

FIGURE 4. SEMLHI conceptual framework.

defining the problem until reaching the stage of development

and ending with the results, as explained in the next section.
To develop the HI system, developers follow many
sequence steps, such as design (encode data, define outliers
and clean the data), implement (verification and validation),
maintain defined workflows, structure information, provide
security and privacy, test the performance, and then release
the software applications. Records in most datasets in HI
are weakly structured and non-standardized. To apply ML
to the HI system, a set of patterns must be used by the
algorithm to predict and visualize the ML algorithm and
generate knowledge. The main patterns that were used in
our framework were the geographic location, patient records,
departments and hospitals, surgical history, obstetric history,
family history, habits, immunization, assessment and plan,
and test results. The next section describes the details of our
framework, which is composed of four components.

B. SEMLHI FRAMEWORK
SEMLHI frameworks were specifically geared toward facil-
itating the development of software applications and include
components that facilitate the analysis of a health dataset.
Many users, as illustrated in Figure 3, will work directly FIGURE 5. SEMLHI framework components.
as developers or system analysts with approach frame-
works or indirectly by using the results. Figure 4 summa-
rizes the proposed framework as a conceptual framework, in algorithms, and health informatics data). Figure 5 shows
addition to the mechanism used to interact with the operating how each module interacts with all modules to work as a
system and hardware. framework.
For software engineers, our frameworks interact with oper-
ating system components that were used by the framework, 1) HEALTH INFORMATICS DATA
and all software manages the device hardware with the main In ML, data are essential, and choosing the methods for
system device used by the framework. presenting and visualizing knowledge is the most impor-
Our framework was composed of four components or mod- tant step. Our dataset sample contains ten columns with
ules (software, machine learning model, machine learning 50k rows (cases). To use a dataset on health informatics

23172 VOLUME 8, 2020

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2986259, IEEE Access
M. Moreb et al.: Novel Software Engineering Approach Toward Using Machine Learning for Improving the Efficiency

FIGURE 6. Machine algorithm model components.

data (HID) algorithms, a transformation into numerical
features was required. Other data contain missing, dupli-
cate or null values, such as negative ages and extremely
large integers, which could negatively affect the performance
of our ML algorithm. Figure 6 describes the main roles in
detecting the methodologies used in the machine algorithm
model, which are classification, clustering, regression and
reduction.
HID uses data sources and a dictionary for translation
during label encoding to convert each value in a column
to a number to reduce the amount of misinterpreted data
used by Bayesian inference. A node identifier was used to
analyze data as a common process with patterns determined
using patient-specific research identifiers. A dataset usually
requires multiple records from the same patient to be identi-
fied as being related in the deidentified database. For outlier
HID, a set of methods was used in the analysis to find hidden
groups to remove outliers, and in an advanced step, the outlier
values of the data that appear to be erroneous need to be found
and cropped from the dataset. Addressing incomplete data in
unsupervised clustering, chi-square and Fisher’s exact tests
were performed to determine the patterns that are discrimi-
nating between pair clusters [29].
To predict disease, we used ICD-10 with multiple labels,
as each patient has an ICD code in their health records,
which can affect all regions of the retina. However, there is
currently no classification system [30] for distinguish-
ing anterior (peripheral) and posterior (macular) data.
We hypothesize that these classifications were characterized
by D and refractive features, highlighting the disparity in the
types of disease.
Collected electrocardiograph data were used to focus on
the D most common diagnosis cases in the laboratory test
result database: D= {d1, d2, d3, . . . , dn }, where d is a disease
that was applicable to a diagnosis code and n is the number
of disease classes using the k-means algorithm with multiple
labels. Algorithm 1 presents the pseudocode for the k-nearest
neighbor algorithm for multilevel learning.

VOLUME 8, 2020 23173

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2986259, IEEE Access
M. Moreb et al.: Novel Software Engineering Approach Toward Using Machine Learning for Improving the Efficiency

FIGURE 7. Age and category feature summary with other variables (test results, gender, and ICD-10).

Algorithm 1 k-Nearest Neighbor Algorithm for Multi-
level Learning by Using Correlation, Diagnosis Code, and
Label Weight With Frequency
input: Heterogeneous data source, number of k,
correlation n. training set;
label of x
while While condition do
1- for all heterogeneous datasets, we will work on
correlating multiple labels, adding one label for i = l
to t and then joining the table for i = l to t, join table
2- apply ML to one label based on Freq. DG weight,
and Lowes accuracy
3- create new role based on step 2
4- apply new role to all new schema; create new role
if micro = sen, and if test = normal, then mml = 3
5- classify ML based on DC category
6. predict new disease based on role created;
end
The data module reads the data from data sources, such as
CSV files or any other available sources; this module includes
a set of algorithms that automatically remove missing values,
clean the data to remove noise, and encode some features.
Predicting missing values with incomplete data, for classifi-
cation, normally requires decision trees; for a small amount
of sample data or large numbers of genes, feature selection
techniques, such as genetic algorithms and particle swarm
optimization, are widely used [31]. The DRFLLS tool gave
the best estimation to estimate missing values for a dataset
that has a small rate of missing values [32].
As we have 750 case categories in our sample test data,
represented by a 27-laboratory test, after running this module,
a new dataset that includes 18 columns and 750 rows is gen-
erated. Figure 7 summarizes the ages with category features
clustered by laboratory test results.
2) ML ALGORITHMS
Machine learning algorithms (MLAs) are used to compute
the parameters that might define a model [14], optimize
its network topology and improve the system convergence
without losing information. MLAs including submodules are
listed in Table 3.
As a supervised learning method, k-nearest neighbors
(KNN) [33] can be used for classification and prediction
problems. KNN makes decisions based on the dominant
categories of k objects rather than a single object category.
Figure 8 identifies most of the MLAs used for health classi-
fication.
As all the data in our sample of datasets were prepared
using the SEMLHI framework, the output method will super-
vise the ‘‘label data’’ for this KNN algorithm with multiple
labels and evaluate our result (KNN was used for super-
vised learning, while k-means was used for unsupervised
learning). k-Means can be used for datasets that include a
million labeled data points. Approximate nearest neighbors

23174 VOLUME 8, 2020

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2986259, IEEE Access
M. Moreb et al.: Novel Software Engineering Approach Toward Using Machine Learning for Improving the Efficiency

TABLE 3. Machine learning algorithms sub model.

FIGURE 9. High accuracy of logistic regression compared with that of

other algorithms.

FIGURE 10. Mechanism of the machine algorithm model.

analysis, dimensional reduction is applied [34]. The training

FIGURE 8. Machine learning algorithms used for health classification.
(ANNs), which is usually 10x - 100x faster than KNN support
vector machines (SVMs), is a good and fast solution for
many problems and will almost always outperform KNNs.
Figure 9 shows that logistic regression has high accuracy
compared with expected and real predictions.
In supervised learning, the dataset contains ‘n’ rows
(cases); each case needs to be evaluated using a function
f : A → B to compare with label A or label B according to
the function ‘f ’ by evaluating E and comparing them to learn
from the training set of n. f has a set n(d). In unsupervised
learning, the data are not labeled. To apply the data in the
set (t) is used, which includes n objects (t) = xi ∈ A: 1
≤ i ≤ 0 that can be in category n of class C1, ck ∈ A,
by applying algorithm f in the evaluation phase to the set,
where ck takes the input x ∈ Cj: 1 ≤ j ≤ k.
For clustering, we need to calculate the distance d between
the two objects x and y by comparing the values of their n
features and applying the Minkowski metric.
3) MACHINE ALGORITHM MODEL
Machine learning helps us extract useful features from a
dataset to address or predict health-related events [35].
The machine algorithm model (MAM) component includes
five submodules: read the data, prepare the data, train the
model, test and evaluate the model, and predict new data.
Figure 10 describes the sequence of these stages.
The challenge for this component was to use the right type
of algorithm, which can optimally solve the dataset while
avoiding high bias or variance. The main component of the
MAM was used to analyze the dataset based on the set of
conditions. If the dataset includes > 50 labeled samples,
then classification algorithms will be used for the selection;

VOLUME 8, 2020 23175

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2986259, IEEE Access
M. Moreb et al.: Novel Software Engineering Approach Toward Using Machine Learning for Improving the Efficiency
TABLE 4. Evaluation of the accuracy of machine learning models.
FIGURE 11. The software module includes subclasses including reuse,
performance, testing, privacy, and security.
otherwise, cluster algorithms will be applied. If the dataset
requires prediction of the quantity, regression algorithms will
be used; otherwise, dimensional reduction will be applied.
Based on the original data, five algorithms were used to
predict the laboratory test results utilizing the MAM com-
ponent of the SEMLHI framework. ML approaches and
algorithms can achieve better performances than expert-
knowledge-based approaches [30]. ML algorithms use two
types of techniques: supervised learning and unsupervised
learning. For the MLA module, we first determine which
techniques to use; then, we select the most suitable algorithms
to use based on mathematical selection related to certain
criteria. For the different algorithms applied, Table 4 shows
their accuracy results (KNN classifier, linear SVC, logistic
regression, multinomial NB, and random forest classifier).
We compared our approach with previously published
systems in terms of performance to evaluate the accuracy
of the machine learning models. The accuracy results for
different algorithms were obtained after applying them to
750 cases, with linear SVC having values of approximately
0.57, compared with the KNN classifier, logistic regression,
multinomial NB, and random forest classifier.
4) SOFTWARE
The software module, which is visualized in Figure 11,
includes a subclass that includes reuse, performance, testing,
privacy, and security. For software testing, the main point
was to verify that the code was running correctly by test-
ing the code under known conditions and checking that the
results were as expected [36]. Visual analytic and interactive
visualizations offer a higher degree of freedom for users
23176
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
for feature filtering, sorting patterns according to different
interestingness measures, templating, and providing details
on demand. Various visualization techniques, such as Even-
tExplorer, ActiviTree, MatrixWave and DecisionFlow, can
be used. Patterns can be clustered using an SOM or projec-
tion method, while plot patterns can use the double-decker
method [37].
This class was used to test the memory or CPU resource
usage for the application. The performance issues were deter-
mined by first measuring them and then profiling the code.
Then, the optimization of that code was carried out using
the benchmark, which was the best choice for comparing the
results to improve the optimization performance. Code smells
found genetic algorithms, used by 22.22%, to be the most
commonly utilized machine learning techniques [38].
In multi-label classification, a prediction containing a sub-
set of the actual classes should be considered better than a
prediction that contains none of them, i.e., predicting two of
the three labels correctly is better than predicting no labels
at all. To measure a multi-class classifier, a misclassification
using micro- and macro-averaging was carried out [16].
The security module has a significant impact on software
development, maintenance, cost, and quality; security pro-
cesses are implemented by integrating security activities and
tools in the software development process, utilizing security
requirement management, and providing training for
developers.
IV. CONCLUSION
This article addressed an important HI with ML topic in
software engineering by proposing an efficient new method
approach related to software engineering, identified in prior
research studies, using original data sets collected during
the last 3 years from a Palestine hospital. This methodology
allows developers to analyze and develop software for the HI
model and create a space in which software engineering and
ML experts can work together on the ML model life-cycle,
especially in the health field. This manuscript proposed a
framework that included a theoretical framework composed
of four modules (software, ML model, ML algorithms, and
HI data). The new methodology was compared between three
system engineering methods: Vee, Agile and SEMLHI. The
results showed the delivery of the new methodology for one-
shot delivery. For the MAM component on the SEMLHI
framework, laboratory test results were obtained using five
algorithms to test the accuracy of the ICD-10 results using
equations and to evaluate the accuracy of the ML models with
a sample size of 750 patients. The results for MAM showed
that the SVC was approximately 0.57.
AVAILABILITY OF DATA AND MATERIALS
Data that support the findings of this research were available
from The Palestinian Ministry of Health, but restrictions were
applied to the availability of these data, which were used
under license for the current study and thus were not publicly
available. Data are, however, available from the authors upon
VOLUME 8, 2020
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2986259, IEEE Access
M. Moreb et al.: Novel Software Engineering Approach Toward Using Machine Learning for Improving the Efficiency
reasonable request and with permission of the Correspon-
dence Author.
ABBREVIATIONS
SEMLHI: Software Engineering for Machine Learning in
Health Informatics
SE: Software Engineering
ML: Machine Learning
HID: Health Informatics Data
HI: Health Informatics
MAM: Machine Algorithm Model
ETHICS APPROVAL AND CONSENT TO PARTICIPATE
The research meets all applicable standards with regard to the
ethics of experimentation and research integrity, and the fol-
lowing was certified/declared true. The informed consent of
human participants was obtained in written format, and it was
approved by The Palestinian Ministry of Health. As an expert
scientist and along with coauthors in the concerned field,
the paper has been submitted with full responsibility, follow-
ing due ethical procedure, and there is no duplicate publica-
tion, fraud, plagiarism, or concerns about animal or human
experimentation.
CONSENT FOR PUBLICATION
Not applicable.
COMPETING INTERESTS
All authors report no conflicts of interest.
FUNDING
Not applicable.
ACKNOWLEDGMENT
The authors would like to thank Health Minister of the State
of Palestine, Dr. J. Awwad, for allowing us to access the
Palestinian dataset for patients, and for all the teams that
supported us during the last two years, the feedback from
whom greatly improved this manuscript.
REFERENCES
[1] A. Holzinger, ‘‘Interactive machine learning: Experimental evidence
for the human in the algorithmic loop,’’ Appl. Intell., vol. 49, no. 7,
pp. 2401–2414, 2019.
[2] T. A. Mohammed, A. Ghareeb, H. Al-Bayaty, and S. Aljawarneh, ‘‘Big
data challenges and achievements: Applications on smart cities and energy
sector,’’ in Proc. 2nd Int. Conf. Data Sci., E-Learn. Inf. Syst., 2019, p. 26.
[3] B. Cakici, K. Hebing, M. Grünewald, P. Saretok, and A. Hulth, ‘‘CASE:
A framework for computer supported outbreak detection,’’ BMC Med.
Inform. Decis. Making, vol. 10, no. 1, p. 14, 2010.
[4] A. J. Vickers, T. Salz, E. Basch, M. R. Cooperberg, P. R. Carroll, F. Tighe,
and J. Eastham, and R. C. Rosen, ‘‘Electronic patient self-assessment and
management (SAM): A novel framework for cancer survivorship,’’ BMC
Med. Inform. Decis. Making, vol. 10, no. 1, p. 34, 2010.
[5] A. Ismail, A. Shehab, and I. M. El-Henawy, ‘‘Healthcare analysis in
smart big data analytics: Reviews, challenges and recommendations,’’ in
Security in Smart Cities: Models, Applications, and Challenges, vol. 9, A.
E. Hassanien, M. Elhoseny, S. H. Ahmed, and A. K. Singh, Eds. Cham,
Switzerland: Springer, Nov. 2019, pp. 27–45.
VOLUME 8, 2020
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
[6] J. F. Bobb, B. C. Henn, L. Valeri, and B. A. Coull, ‘‘Statistical software for
analyzing the health effects of multiple concurrent exposures via Bayesian
kernel machine regression,’’ Environ. Health, vol. 17, no. 1, p. 67, 2018.
[7] B. Aribisala and O. Olabanjo, ‘‘Medical image processor and repository–
MIPAR,’’ Inform. Med. Unlocked, vol. 12, pp. 75–80, Jul. 2018.
[8] W. Aigner and S. Miksch, ‘‘CareVis: Integrated visualization of comput-
erized protocols and temporal patient data,’’ Artif. Intell.in Med., vol. 37,
no. 3, pp. 203–218, Jul. 2006.
[9] J. Krause, A. Perer, and H. Stavropoulos, ‘‘Supporting iterative cohort
construction with visual temporal queries,’’ IEEE Trans. Vis. Comput.
Graph., vol. 22, no. 1, pp. 91–100, Jan. 2016.
[10] R. K. Pathinarupothi, P. Durga, and E. S. Rangan, ‘‘Data to diagnosis in
global health: A 3P approach,’’ BMC Med. Inform. Decis. Making, vol. 18,
no. 1, pp. 1–13, 2018.
[11] I. H. Witten, E. Frank, M. A. Hall, and C. J. Pal, Data Mining: Practical
Machine Learning Tools and Techniques. Amsterdam, The Netherlands:
Elsevier, 2016, pp. 438–441.
[12] Q.-C. To, J. Soto, and V. Markl, ‘‘A survey of state management in big data
processing systems,’’ VLDB J., vol. 27, no. 6, pp. 847–872, Dec. 2018.
[13] S. R. Salkuti, ‘‘A survey of big data and machine learning,’’ Int. J. Elect.
Comput. Eng., to be published. Accessed: Jan. 7, 2020. [Online]. Available:
http://ijece.iaescore.com/index.php/IJECE/article/view/19184/pdf
[14] F. Khomh, B. Adams, J. Cheng, M. Fokaefs, and G. Antoniol, ‘‘Software
engineering for machine-learning applications: The road ahead,’’ IEEE
Softw., vol. 35, no. 5, pp. 81–84, Sep. 2018.
[15] T. A. Mohammed, Y. I. Hamodi, and N. T. Yousir, ‘‘Intelligent enhance-
ment of organization work flow and work scheduling using machine learn-
ing approach tree algorithm,’’ Int. J. Comput. Sci. Netw. Secur., vol. 18,
no. 6, pp. 87–90, 2018.
[16] J. A. Diao, I. S. Kohane, and A. K. Manrai, ‘‘Biomedical informatics
and machine learning for clinical genomics,’’ Hum. Mol. Genet., vol. 27,
no. R1, pp. R29–R34, May 2018.
[17] P.-H. Cheng, Y.-P. Chen, and J.-S. Lai, ‘‘An interflow system requirement
analysis in health informatics field,’’ in Proc. WRI World Congr. Comput.
Sci. Inf. Eng., vol. 1, 2009, pp. 712–716.
[18] C. George, P. Duquenoy, and D. Whitehouse, ‘‘eHealth: Legal, ethical
and governance challenges,’’ in eHealth: Legal, Ethical and Governance
Challenges, C. George, D. Whitehouse, and P. Duquenoy, Eds. Berlin,
Germany: Springer, 2014, pp. 1–398.
[19] K. N. Mishra and C. Chakraborty, ‘‘A novel approach towards using
big data and IoT for improving the efficiency of m-health systems,’’ in
Advanced Computational Intelligence Techniques for Virtual Reality in
Healthcare, vol. 875. Cham, Switzerland: Springer, 2020, pp. 123–139.
[20] B. Farahani, M. Barzegari, F. Shams Aliee, and K. A. Shaik, ‘‘Towards
collaborative intelligent IoT eHealth: From device to fog, and cloud,’’
Microprocessors Microsyst., vol. 72, Feb. 2020, Art. no. 102938.
[21] C. Oliver, ‘‘Critical realist grounded theory: A new approach for social
work research,’’ Brit. J. Social Work, vol. 42, no. 2, pp. 371–387, Mar. 2012.
[22] J. Disantostefano, ‘‘International classification of diseases 10th revision
(ICD-10),’’ J. Nurse Practitioners, vol. 5, no. 1, pp. 56–57, Jan. 2009.
[23] K. D. Clark, T. T. Woodson, R. J. Holden, R. Gunn, and D. J. Cohen,
‘‘Translating research into agile development (TRIAD): Development of
electronic health record tools for primary care settings,’’ Methods Inf. Med.,
vol. 58, no. 1, pp. 1–8, Jun. 2019.
[24] T. A. Mohammed, S. Alhayli, S. Albawi, and A. Deniz Duru, ‘‘Intelligent
database interface techniques using semantic coordination,’’ in Proc. 1st
Int. Sci. Conf. Eng. Sci.-3rd Sci. Conf. Eng. Sci. (ISCES), Jan. 2018,
pp. 13–17.
[25] T. A. Mohammed, O. Bayat, O. N. Uçan, and S. Alhayali, ‘‘Hybrid Effi-
cient Genetic Algorithm for Big Data Feature Selection Problems,’’ Found.
Sci., to be published.
[26] T. A. Mohammed, S. Alhayali, O. Bayat, and O. N. Uçan, ‘‘Feature
reduction based on hybrid efficient weighted gene genetic algorithms with
artificial neural network for machine learning problems in the big data,’’
Sci. Program., vol. 2018, pp. 1–10, Oct. 2018.
[27] T. Weilkiens, J. G. Lamm, S. Roth, and M. Walker, ‘‘B: The V-Model,’’
in Model-Based System Architecture. Hoboken, NJ, USA: Wiley, 2015,
pp. 343–352.
[28] M. Al-Zewairi, M. Biltawi, W. Etaiwi, and A. Shaout, ‘‘Agile software
development methodologies: Survey of surveys,’’ J. Comput. Commun.,
vol. 05, no. 05, pp. 74–97, 2017.
[29] Y. Zhou, ‘‘Predictive big data analytics using the UK Biobank data,’’ Sci.
Rep., vol. 9, no. 1, p. 6012, Dec. 2019.
23177
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2986259, IEEE Access
M. Moreb et al.: Novel Software Engineering Approach Toward Using Machine Learning for Improving the Efficiency

[30] A. J. Steele, S. C. Denaxas, A. D. Shah, H. Hemingway, and TAREQ ABED MOHAMMED received the B.Sc.
N. M. Luscombe, ‘‘Machine learning models in electronic health records degree in computer science from the College
can outperform conventional survival models for predicting patient mor- of Science, Kirkuk University, Kirkuk, Iraq,
tality in coronary artery disease,’’ PLoS ONE, vol. 13, no. 8, Aug. 2018, in 2007, the M.Sc. degree from Cankaya Univer-
Art. no. e0202344. sity, Ankara, Turkey, in 2012, and the Ph.D. degree
[31] W. Pearson, C. T. Tran, M. Zhang, and B. Xue, ‘‘Multi-round random in electronic and computer engineering from
subspace feature selection for incomplete gene expression data,’’ in Proc. Altinbas University, Istanbul, Turkey, in 2019.
IEEE Congr. Evol. Comput. (CEC), Jun. 2019, pp. 2544–2551.
In 2019, he started teaching at the College of
[32] S. Al-Janabi and A. F. Alkaim, ‘‘A nifty collaborative analysis to predicting
Computer Science and Information Technology,
a novel tool (DRFLLS) for missing values estimation,’’ Soft Comput.,
vol. 24, no. 1, pp. 555–569, Jan. 2020. University of Kirkuk. He has advised many studies
[33] J. Salvador-Meneses, Z. Ruiz-Chavez, and J. Garcia-Rodriguez, ‘‘Com- for M.Sc. and Ph.D. students at various universities, participated in many
pressed kNN: K-nearest neighbors with data compression,’’ Entropy, international conferences and contributed to various scientific studies.
vol. 21, no. 3, p. 234, Mar. 2019. OGUZ BAYAT received the B.S. degree from
[34] D. A. Clifton, J. Gibbons, J. Davies, and L. Tarassenko, ‘‘Machine learning Istanbul Technical University, Istanbul, Turkey,
and software engineering in health informatics,’’ in Proc. 1st Int. Workshop in 2000, the M.S degree from the University of
Realizing AI Synergies Softw. Eng. (RAISE), Jun. 2012, pp. 37–41. Hartford, CT, USA, in 2002, and the Ph.D. degree
[35] E. Boonchieng and K. Duangchaemkarn, ‘‘Digital disease detection: from Northeastern University, Boston, MA, USA,
Application of machine learning in community health informatics,’’ in
in 2006, all in electrical engineering.
Proc. 13th Int. Joint Conf. Comput. Sci. Softw. Eng. (JCSSE), Jul. 2016.
He completed the Executive Certificate Pro-
[36] J. Frochte and J. Frochte, ‘‘Python, NumPy, SciPy und Matplotlib—In a
nutshell,’’ in Maschinelles Lernen. Munich, Germany: Carl Hanser Verlag
gram in Technical Management and Leadership
GmbH, 2019, pp. 32–67. at Massachusetts Institute of Technology, Boston,
[37] W. Jentner and D. A. Keim, ‘‘Visualization and visual analytic tech- MA, USA, in 2009. Since 2011, he has been
niques for patterns,’’ in High-Utility Pattern Mining. Cham, Switzerland: serving as a Professor with the Department of Electrical and Electronics
Springer, 2019, pp. 303–337. Engineering, Altinbas University. He is also an Advisor to the President and
[38] M. I. Azeem, F. Palomba, L. Shi, and Q. Wang, ‘‘Machine learning tech- the Director of the Graduate School of Science and Engineering, Altinbas
niques for code smell detection: A systematic literature review and meta- University.
analysis,’’ Inf. Softw. Technol., vol. 108, pp. 115–138, Apr. 2019. OGUZ ATA received the BSc degree
Computer Engineering from Sakarya
University in 2004, and the MSc degrees
MOHAMMED MOREB was born in Hebron, Computer Engineering from Beykent
Palestine, in 1981. He received the B.Sc. degree in University in 2008, and PhD degrees in
information technology from Palestine Polytech- software engineering from Trakya
nic University, and the M.Sc. degree in computer Üniversitesi in 2012. He has been the head
science from Al-Quds University. He is currently of Department at Software Engineering at
pursuing the Ph.D. degree in electronic and com- Altinbas University and lecturer at
puter engineering with Altinbas University. Altinbas University. His research interests
The focus of his research is software engineer- include software repository mining, software measurement and testing,
ing in health informatics. He has over twelve years process improvement, and requirements engineering
of experience in managing software development
projects, including large government IT systems.

23178 VOLUME 8, 2020

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.